PBPK-Based Digital Twins for Predictive Dosimetry In PSMA Radiopharmaceutical Therapy: A Virtual Theranostic Trial Study
Abstract
Purpose
Existing "one-size-fits-all" approaches in PSMA radiopharmaceutical therapies (RPTs) fail to account for critical inter-patient variabilities, risking under- or over-treatment. To address this, we are developing theranostic digital twins (TDTs) for reliable predictive dosimetry towards precision RPTs. This study utilizes an in silico virtual theranostic trials (VTTs) framework to evaluate whether TDTs, constructed via physiologically based pharmacokinetic (PBPK) modeling, can accurately predict therapeutic dosimetry using pre-therapy diagnostic PET scans.
Methods
We conducted a VTT with 50 virtual patients, utilizing a knowledge-driven PBPK model including tumor, salivary glands (SG) and kidneys. Literature-based pharmacokinetic (PK) parameters were used to generated 18F- and 64Cu-PSMA PET imaging with realistic noise under two acquisition scenarios: (1) standard early dynamic PET (0–65 min) alone; and (2) early dynamic PET supplemented with two 10-minute discrete late-window static frames (4 and 8 hours for 18F-PSMA; 24 and 48 hours post-injection for 64Cu-PSMA). A Cluster Gauss-Newton algorithm estimated 10 patient-specific parameters: blood flow fractions (f), receptor densities (Rden), and release rates λrel for the tumor, SG, and kidneys, alongside kidney excretion rate (fexc). Predicted RPT doses were assessed via percentage error (PE: mean±std).
Results
Early-window imaging alone resulted in large dosimetric deviations. For 18F, PE% values were 36.3±26.2% (SG), 54.7±40.5% (kidney), and -14.9±52.2% (tumor). Similarly, 64Cu early-window imaging yielded 49.9±37.2% (SG), 58.7±40.5% (kidney), and -5.5±74.3% (tumor). Incorporating late static frames significantly improved accuracy. For 18F (4h/8h scans), PEs reduced to 7.5±13.7% (SG), 15.0±15.5% (kidney), and 25.7±34.7% (tumor). 64Cu protocol (24h/48h scans) achieved the highest precision, reducing errors to 9.2±8.3% (SG), 15.9±6.9% (kidney), and 11.4±19.0% (tumor).
Conclusion
Standard early-window PET is insufficient for reliable TDT-based dosimetry due to poor parameter identifiability. Incorporating late-time static frames is essential for accurate therapeutic planning. This VTT framework successfully optimizes acquisition protocols in silico, highlighting the superior quantitative potential of 64Cu-PSMA with delayed imaging.